The defect detection in glass materials by using discrete wavelet packet transform and artificial neural network

Abstract

In this study, a method based on impact tests was designed in order to determine undamaged and broken glasses. By means of using an impact pendulum, impact was applied on glasses and the generated sounds were transferred to the computer using a microphone. The sound signals were decomposed into 128 components by using Discrete Wavelet Packet Transform (DWPT) at the seventh level. 16 of the 128 components that characterized the properties of undamaged and broken glasses were chosen as inputs for the designed Artificial Neural Network (ANN). The designed ANN model was tested with real-time simulation, and it was observed that the proposed method could determine undamaged and broken glasses with high precision. This method, which is based on analyzing the sounds generated after the impact, can detect defects that the conventional visual methods can detect; however, it can also be used as supplement to these methods.